hive-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From David Ginzburg <ginz...@hotmail.com>
Subject RE: inconsistent results when doing a select over a join
Date Tue, 10 Jan 2012 15:44:58 GMT


Adding more input 

The input size of the query was about 353Gb resulted in 638 reducers on a 70 node cluster
with Reduce Task Capacity of 286.



> Date: Tue, 10 Jan 2012 17:19:59 +0200
> From: guy.doulberg@conduit.com
> To: user@hive.apache.org
> CC: edlinuxguru@gmail.com
> Subject: Re: inconsistent results when doing a select over a join
> 
> Hi guys,
> I spent the day today investigating this issue, it seems like the 
> differences occur when there are many killed tasks.
> 
> We are using the fair scheduler,  I ran the queries on large data and 
> with low priority which caused the tasks of this job to be 
> preempt(killed) many times.
> 
> After I began suspecting this issue, I gave the query the highest 
> priority by doing that I reduced the number of killed tasks, that 
> seemed to solve the problem
> 
> 
> It is not that whenever there are killed task there are differences, it 
> is when there many killed task because of preemption there are 
> differences.
> 
> What do you say?
>     
> On Tue 10 Jan 2012 11:49:35 AM IST, Guy Doulberg wrote:
> > Hi,
> > Sorry for the late answer,
> > I ran the query on small data, but couldn't reproduce,
> > I can reproduce it at the moment on data that takes about 1.5  hour to
> > process,
> > I am trying to narrow the amount of data as much as I can, and still
> > reproduce it...
> >
> > But I think it is clear to me, that the scale of data is the reason for
> > the differences,
> >
> > What do you think?
> >
> >
> >
> > On Mon 09 Jan 2012 08:14:10 PM IST, Edward Capriolo wrote:
> >> Create table, query , and some small data set to reproduce
> >>
> >> On Monday, January 9, 2012, Guy Doulberg<guy.doulberg@conduit.com
> >> <mailto:guy.doulberg@conduit.com>>  wrote:
> >>> Thanks, I am trying to reproduce it again,
> >>>
> >>> But what should I send the ML?
> >>>
> >>>
> >>>
> >>>
> >>> On Mon 09 Jan 2012 07:54:24 PM IST, Edward Capriolo wrote:
> >>>>
> >>>> Can you reproduce the issue? possibly with the smaller tables and
> >> send that to the ML?
> >>>>
> >>>> Edward
> >>>>
> >>>> On Mon, Jan 9, 2012 at 12:46 PM, Guy Doulberg
> >> <guy.doulberg@conduit.com<mailto:guy.doulberg@conduit.com>
> >> <mailto:guy.doulberg@conduit.com<mailto:guy.doulberg@conduit.com>>>
> >> wrote:
> >>>>
> >>>>     Hey Dave,
> >>>>     I didn't understand your question,
> >>>>
> >>>>     The Inconsistant is slightly different, about 2% of differences,
> >>>>
> >>>>     Thanks
> >>>>
> >>>>     Guy
> >>>>
> >>>>     On 01/09/2012 07:05 PM, David Houston wrote:
> >>>>>
> >>>>>     Hi Guy,
> >>>>>
> >>>>>     Inconsistant by way of the results are total off or the order
is
> >>>>>     different?
> >>>>>
> >>>>>     Thanks
> >>>>>
> >>>>>     Dave
> >>>>>
> >>>>>     On Jan 9, 2012 5:03 PM, "Guy Doulberg"
> >> <guy.doulberg@conduit.com<mailto:guy.doulberg@conduit.com>
> >>>>> <mailto:guy.doulberg@conduit.com
> >> <mailto:guy.doulberg@conduit.com>>>  wrote:
> >>>>>
> >>>>>         Hi guys,
> >>>>>
> >>>>>         We are using hive for a while now, and recently we have
> >>>>>         encountered an issue we just can't understand,
> >>>>>
> >>>>>         We are selecting(the select includes count(*)) over a join
of
> >>>>>         two big tables.
> >>>>>
> >>>>>         We ran the same query twice consequently over the same two
> >>>>>         tables , and each time the result were slightly different.
> >>>>>
> >>>>>         We don't know how should we debug this issue, where should
we
> >>>>>         look, any ideas?
> >>>>>
> >>>>>         Thanks
> >>>>>
> >>>>>         Guy Doulberg,
> >>>>>         Data infrastructure engineer,
> >>>>>         Conduit
> >>>>>
> >>>>
> >>>
 		 	   		  
Mime
View raw message